AltOverview Psnup
AltOverview Psnup
AltOverview Psnup
Todays manufacturers face strong pressure to: Improve productivity, product eld reliability, and overall quality using new technology. Develop newer, higher technology products in record time. Implies increased need for up-front testing of materials, components and systems. Accelerated tests provide timely information for product design and development. Users must be aware of potential pitfalls
William Q. Meeker Department of Statistics and Center for Nondestructive Evaluation Iowa State University Ames, IA 50011
Overview What is Reliability? Dierent kinds of accelerated tests R(t) = 1 F (t) The probability that a system, vehicle, machine, device, and so on, will perform its intended function under encountered operating conditions, for a specied period of time. Quality over time A powerful marketing tool An engineering discipline requiring support from Physics and chemistry Statistics Example 1Evaluation of an insulating structure Example 2New-technology microelectronic logic device Accelerated Degradation Tests Importance of physics of failure and physical/chemical models (and sensitivity analysis) Example 3Microelectronic RF amplier device Connecting with the eld Example 4Appliance eld reliability Areas for further research
Breakdown Times in Minutes of a Mylar-Polyurethane Insulating Structure (from Kalkanis and Rosso 1989)
Assess component or material reliability or durability. Make design decisions to improve reliability or lower cost Verify predictions produced with physical models (e.g. FEM)
Minutes
10
10
System test to simulate eld-use at accelerated conditions. Predict product eld performance. Identify and x potential failure modes at system/subsystem level (HALT and STRIFE tests). Screening (100% or audit) testing of manufactured product (e.g. ESS and burn-in).
10
10
10
10
-1
100
150
200 kV/mm
250
300
350 400
Plot of Inverse Power Relationship-Lognormal Model Fitted to the Mylar-Polyurethane Data (also Showing 361.4kV Data Omitted from the ML Estimation)
log(t)
10 10
Minutes
10 10 10 10
-1
50
100 kV/mm
200
500
Lognormal Probability Plot of the Inverse Power Relationship-Lognormal Model Fitted to the Mylar-Polyurethane Data
Proportion Failing
Increase the use-rate of the product (e.g., test a toaster 400 times/day). Higher use rate reduces test time. Use elevated temperature or humidity to increase rate of failure-causing chemical/physical process. Increase stress (e.g., voltage or pressure) to make degrading units fail more quickly. Use a physical/chemical (preferable) or empirical model relating degradation or lifetime at use conditions.
10
5
219.0
1
157.1
122.4
2
100.3
3
50 kV/mm
4
10
10
10
10
Minutes
10
Interval ALT Data for a New-Technology IC Device Tests run at 150, 175, 200, 250, and 300C.
10
Developers interested in estimating activation energy of the suspected failure mode and the long-life reliability.
Hours
10
10
Failures had been found only at the two higher temperatures. After early failures at 250 and 300C, there was some concern that no failures would be observed at 175C before decision time. Thus the 200C test was started later than the others.
10
10
x x x
x x x
10
100
150
200 Degrees C
250
300
350
11
12
The Arrhenius-Lognormal Regression Model Elevated Temperature Acceleration of Chemical Reaction Rates The Arrhenius model Reaction Rate, R(temp), is
Ea Ea 11605 = 0 exp kB (temp C + 273.15) temp K where temp K = temp C+273.15 is temperature in Kelvin and R(temp) = 0 exp
kB = 1/11605 is Boltzmanns constant in units of electron volts per K. The reaction activation energy, Ea, and 0 are characteristics of the product or material being tested. The reaction rate Acceleration Factor is
AF(temp, tempU , Ea ) = R(temp) R(tempU ) 11605 11605 tempU K temp K
= exp Ea
13
14
Arrhenius Plot Showing ALT Data and the Arrhenius-Lognormal Model ML Estimation Results for the New-Technology IC Device.
Lognormal Probability Plot Showing the Arrhenius-Lognormal Model ML Estimation Results for the New-Technology IC Device
.95
10
.9 .8
10
.7 .6 .5 .4 .3 .2 .1 .05 .02
Hours
10
10
Proportion Failing
10
x x x
x x x
10
250 10
3
200 10
4
175
150 10
5
100 10
6
100
150
200
250
300
350
10
Hours
16
Lognormal Probability Plot Showing the Arrhenius-Lognormal Model ML Estimation Results for the New-Technology IC Device with Given Ea = .8 Pitfall 4: Masked Failure Mode
.95 .9 .8 .7 .6 .5 .4 .3 .2 .1 .05 .02 .01 .005 .002 .0005 .0001 10 300 Deg C
2
Accelerated test may focus on one known failure mode, masking another!
Proportion Failing
Masked failure modes may be the rst one to show up in the eld.
200 10
4
175
150 10
5
100 10
6
10
Hours
17
18
10
10
10
10
Hours
Hours
10
10
4
Mode 2
10
10
10%
10
2
10%
10
Mode 1 10%
10
40
60
80 Degrees C
100
120
140
19
20
Percent Increase in Resistance Over Time for Carbon-Film Resistors (Shiomi and Yanagisawa 1979)
5.0
Percent Increase
Can be more informative than time-to-failure data. (Reduction to failure-time data loses information)
0.5
Hours
21 22
Degradation data may be dicult or impossible to obtain (e.g., destructive measurements). Obtaining degradation data may have an eect on future product degradation (e.g., taking apart a motor to measure wear). Substantial measurement error can diminish the information in degradation data. Analyses more complicated; requires statistical methods not yet widely available. (Modern computing capabilities should help here) Degradation level may not correlate well with failure.
Percent Increase in Operating Current
0
0
10
15
1000
2000 Hours
3000
4000
23
24
Device-B Power Drop Accelerated Degradation Test Results at 150C, 195C, and 237C (Use conditions 80C)
A1 k1 -A2
and the rate equations for this reaction are
0.0
-0.2
Power drop in dB
-0.4
-0.6
-0.8
-1.0
-1.2
-1.4
and
dA2 = k1A1, dt
k1 > 0.
(1)
k1 = 0 exp
Ea kB (temp + 273.15)
26
Hours
25
Lognormal-Arrhenius Model Fit to the Device-B Time-to-Failure Data with Degradation Model Estimates
What Do Accelerated Test Results Tell Us About Field Reliability? Need information on:
Eects of acceleration (e.g., cycling rate). Distribution of use-rates in actual use. Distribution of environmental conditions (e.g., stress spectra distributions).
237 Degrees C 195 150 Degrees C 80 Degrees C
Proportion Failing
.7 .6 .5 .4 .3 .2 .1 .05 .02 .01 .005 .001 10^1 10^2 10^3 10^4 10^5
These factors may be given or, in some situations, inferred from the available data.
Hours
27
28
Carefully compare laboratory tests results and eld failures. Same failure mechanisms operating in laboratory tests? Same factors (environmental noises) exciting the failure mechanisms? Identify laboratory/eld discrepancies to improve test procedures. Seek understanding of reasons for lack of agreement. Find a model (transfer function) to relate laboratory test to eld use. Understanding the relationship between the laboratory test results and product eld reliability will provide stronger basis for using future laboratory tests to predict eld performance.
10000
20000 Cycles
30000
40000
50000
29
30
log(c)
Actual use-rate has a distribution given by the proportion of users i (i = 1, . . . , k) that use the appliance at constant rate Ri, where k i = 1. i=1 Then the failure probability as a function of time is FT (t; ) = P (T t) =
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Relative Frequency of Appliance Uses per Week
0.05
0.10
0.0
i
i=1
log (t) i
31
32
.999 .98 .9 .7 .5 .3 .2
.95 .9 .8 .7 .6 .5
Fraction Failing
.1
Probability
50
100
200
1000
2000
5000
Weeks of Service
33
34
Field Variability Lognormal f (r; R, R ) Density for the Wear Failure Mode (unloaded cycles relative to eld days of use)
Lab: subset AccWear Appliance B Wear Failure Mode ALT data Field: subset Field Appliance B Wear Failure Mode
Fraction Failing
0.01
0.05
0.20
1.00
5.00
20.00
In censored accelerated life tests (failure time is response) allocate more test units to low acceleration factor level than high acceleration factor levels.
10
Days
10
Consider including some tests at the use conditions. Use simulation to investigate properties of alternative ALT plans.
10
1
10% Results based on 500 simulations Lines shown for 50 simulations
10
40
60
80
100
120
140
160
Degrees C
37
38
Concluding Remarks
Physical/statistical models for failure acceleration Methods for sensitivity analysis when empirical models must be used Prediction of service life in complicated environments Physical/statistical models the eld environment Bayesian methods for analysis and planning (especially adaptive test plans) Accelerated degradation test planning Degradation analysis and planning with coarse (e.g. ordered categorical and censored) data. Physical comparison of lab and led failures to validate testing methods
39
Accelerated Testing can be valuable tool when used carefully There is no magic in Accelerated Testing Cross-disciplinary teams are needed to deal eectively with all issues Product/reliability/design engineers to identify productuse proles, environmental considerations, potential failure modes or weaknesses that need to be evaluated, etc. Experts in materials and the chemistry/physics of failure to help in the understanding of an suggest/develop appropriate models for acceleration of particular failure modes. Statisticians to help with stochastic modeling, plan tests, t models, and to help quantify uncertainty in results. Users of Accelerated Testing must beware of pitfalls
40
References References D. Byrne, J. Quinlan, Robust function for attaining high reliability at low cost, 1993 Proceedings Annual Reliability and Maintainability Symposium, 1993, pp 183-191. L. W. Condra, Reliability Improvement with Design of Experiments, 1993, New York: Marcel Dekker, Inc. M. Hamada, Using statistically designed experiments to improve reliability and to achieve robust reliability, IEEE Transactions on Reliability R-44, 1995 June. M. Hamada, Analysis of experiments for reliability improvement and robust reliability, in Recent Advances in Life-Testing and Reliability, 1995, N. Balakrishnan, editor, Boca Raton: CRC Press. Meeker, W.Q. and Hamada, M. (1995), Statistical Tools for the Rapid Development & Evaluation of High-Reliability Products, IEEE Transactions on Reliability R-44, 187-198.
41
Meeker, W.Q. and Escobar, L.A. (1998a), Statistical Methods for Reliability Data. John Wiley and Sons, Inc. Meeker, W.Q. and Escobar, L.A. (1998b), Pitfalls of Accelerated Testing. , IEEE Transactions on Reliability R-47, 114-118. W. Nelson, Accelerated Testing: Statistical Models, Test Plans, and Data Analyses, 1990, New York: John Wiley & Sons, Inc. G. Taguchi, System of Experimental Design, 1987; White Plains, NY: Unipub/Kraus International Publications. T. S. Tseng, M. Hamada, C. H. Chiao, (1995), Using degradation data from a factorial experiment to improve uorescent lamp reliability, Journal of Quality Technology, 363-369.
42